News · 2026-11-02

Frank Calegari argues mathematics must change how it rewards research

Mathematician Frank Calegari is calling for changes to hiring, journals, graduate education and seminars as AI reduces the cost of producing mathematical results. His November 1 guest essay on Terence Tao’s blog argues that choosing worthwhile questions and understanding their connections deserve more credit. It is an institutional reform argument, not an announcement that AI has solved mathematics.

Key facts

• The essay appeared on November 1, 2026, on Terence Tao’s What’s New.
• Its author is University of Chicago mathematician Frank Calegari, not Tao.
• It addresses four central institutions: hiring, journals, graduate training and seminars.
• The primary source is Calegari’s essay and its hosted discussion.

A research profession built around scarce proofs faces a different problem when generating possible proofs becomes easier. Calegari’s answer is to look beyond the finished result and reward the intellectual work around it: identifying a question worth asking, understanding an answer and building a larger program from it. His university profile places his own work in algebraic number theory and the Langlands program.

His example is deliberately exploratory. He describes using AI to investigate a possible connection between arithmetic hyperbolic three-manifolds and Kolyvagin systems. A reader does not need those technical subjects to understand the proposal. They are distant areas whose prerequisites make a first attempt expensive. Calegari says the idea could be nonsense; the value is lowering the cost of looking across the boundary. He calls the activity “not trying to prove a theorem.”

The distinction resembles using a better map before committing to an expedition. A map can make it cheaper to notice a promising route without proving the route is passable. Calegari wants the profession to recognize the work of selecting and understanding such routes, rather than treating a new destination as the only contribution. His historical examples support an argument about research culture; they are not an experiment showing that the proposed reforms work.

In hiring, he favors assessing technical ability, breadth and a coherent research narrative through direct interviews. A candidate who can explain what makes a program possible may reveal more than a publication count. In journals, he wants higher standards for exposition, ideas and utility, and fewer, better papers. He also raises the possibility of stepping away from journals that cannot adapt. These are proposals, not adopted policies or a consensus position among mathematicians.

For graduate education, the essay emphasizes breadth and durable understanding of how different areas connect. It also preserves the need for deep technical work without AI. Seminars would begin with major questions, blocked approaches and possible directions, reaching the latest result later. The common mechanism is to make the research program legible, rather than presenting each theorem as an isolated trophy.

Formal verification has a supporting role. Calegari suggests that machine-generated results can join the mathematical record, ideally checked in Lean. Ground Truth’s proof-assistant lesson explains the boundary: a kernel checks a proof of a formal statement under an explicit environment. It does not decide whether the statement captures the intended question or whether the result matters. The separately verified code-verification paper illustrates frozen statements and kernel acceptance; it is background, not evidence cited by Calegari.

The strongest objection comes from the essay’s own discussion. Gordon Royle, whose role is documented by his University of Western Australia profile, asks how hiring, tenure and grant committees would screen applicants without publication counts. Interviews may reveal understanding, but they can also reward familiarity, prestigious training and personal networks. Replacing an imperfect count with a more subjective judgment can make access less fair.

Other commenters press a deeper question: will selecting questions and explaining results remain scarce human advantages? If AI becomes strong at those tasks too, and access depends on a few companies, shifting academic rewards upward may not preserve careers or independence. Supportive comments describe AI opening valuable research possibilities, but their accounts are first-person reactions rather than independently checked mathematical results. The Hacker News discussion likewise contains both enthusiasm and concern about employment and credit.

The essay matters because it names concrete institutions that must respond, rather than stopping at a prediction about machines. Its proposals still lack a tested implementation plan for fair hiring, affordable access or governance. The near-term question is whether mathematics can reward understanding and exploration more effectively while preserving rigorous verification and a viable path for younger researchers. Calegari has made that argument explicit; the profession has not yet demonstrated that the answer will work.

Primary source, verified: read the paper →

Ground Truth. (2026, November 2). Frank Calegari argues mathematics must change how it rewards research. Ground Truth. https://groundtruth.day/news/frank-calegari-ai-mathematics-institutions.html

@misc{groundtruth:frank-calegari-ai-mathematics-institutions,
  title  = {Frank Calegari argues mathematics must change how it rewards research},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {nov},
  url    = {https://groundtruth.day/news/frank-calegari-ai-mathematics-institutions.html}
}

Topics: mathematics · ai-and-science · education · research-culture